Short-form single-page QA on NQ (Natural Questions) (official)
50.2EMAGAR
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| AGARBackbone=Qwen3-VL-8B-Instruct2026.06 | 50.2 | 66.5 | 291 | |
| AGLABackbone=Qwen3-VL-8B-Instruct2026.06 | 48.8 | 64.6 | 268 | |
| UniformBackbone=Qwen3-VL-8B-Instruct2026.06 | 48.2 | 64.4 | 292 | |
| Plain VQABackbone=Qwen3-VL-8B-Instruct2026.06 | 47.3 | 63.8 | 268 | |
| VEA-BrBackbone=Qwen3-VL-8B-Instruct2026.06 | 47.2 | 63.9 | 268 | |
| VEA-ConBackbone=Qwen3-VL-8B-Instruct2026.06 | 47.2 | 63.9 | 268 | |
| RandomBackbone=Qwen3-VL-8B-Instruct2026.06 | 47.1 | 63.7 | 301 | |
| VARBackbone=Qwen3-VL-8B-Instruct2026.06 | 46.9 | 63.6 | 268 | |
| CGRBackbone=Qwen3-VL-8B-Instruct2026.06 | 11.5 | 29.3 | 268 |